A Guided Walk Down Wall Street: An Introduction to Econophysics
A Guided Walk Down Wall Street: An Introduction to Econophysics
A Guided Walk Down Wall Street: An Introduction to Econophysics
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
1060 Giovani L. Vasconcelos<br />
where<br />
r = 1 T<br />
T∑<br />
r(t ′ ). (144)<br />
t ′ =1<br />
Next we break up X(t) in<strong>to</strong> N non-overlapping time intervals,<br />
I n , of equal size τ, where n = 0, 1, ..., N − 1 and N<br />
corresponds <strong>to</strong> the integer part of T/τ. We then introduce<br />
the local trend function Y τ (t) defined by<br />
Y τ (t) = a n + b n t for t ∈ I n , (145)<br />
where the coefficients a n and b n represent the least-square<br />
linear fit of X(t) in the interval I n . Finally, we compute the<br />
rescaled fluctuation function F (τ) defined as [35]<br />
F (τ) = 1 √ 1 ∑Nτ<br />
[X(t) − Y τ (t)] 2 , (146)<br />
S nτ<br />
t=1<br />
where S is the data standard deviation<br />
S = √ 1 T∑<br />
(r t − r) 2 . (147)<br />
T<br />
t=1<br />
The Hurst exponent H is then obtained from the scaling behavior<br />
of F (τ):<br />
F (τ) = Cτ H , (148)<br />
where C is a constant independent of the time lag τ.<br />
In a double-logarithmic plot the relationship (148) yields<br />
a straight line whose slope is precisely the exponent H, and<br />
so a linear regression of the empirical F (τ) will immediately<br />
give H. One practical problem with this method, however,<br />
is that the values obtained for H are somewhat dependent<br />
on the choice of the interval within which <strong>to</strong> perform<br />
the linear fit [35, 36]. It is possible <strong>to</strong> avoid part of this difficulty<br />
by relying on the fact that for the fractional Brownian<br />
motion, the fluctuation function F (τ) can be computed exactly<br />
[31]:<br />
F H (τ) = C H τ H , (149)<br />
where<br />
[<br />
2<br />
C H =<br />
2H + 1 + 1<br />
H + 2 − 2 ] 1/2<br />
. (150)<br />
H + 1<br />
In (149) we have added a subscript H <strong>to</strong> the function F <strong>to</strong><br />
denote explicitly that it refers <strong>to</strong> W H (t). Equation (149)<br />
with (150) now gives a one-parameter estima<strong>to</strong>r for the exponent<br />
H: one has simply <strong>to</strong> adjust H so as <strong>to</strong> obtain the<br />
best agreement between the theoretical curve predicted by<br />
F H (τ) and the empirical data for F (τ).<br />
7.3 Fractional Brownian motion in Finance<br />
The idea of using the FMB for modeling asset price dynamics<br />
dates back <strong>to</strong> the work of Mandelbrot & van Ness<br />
[37]. Since then, the Hurst exponent has been calculated<br />
(using different estima<strong>to</strong>rs) for many financial time series,<br />
such as s<strong>to</strong>ck prices, s<strong>to</strong>ck indexes and currency exchange<br />
rates [38, 39, 40, 41, 35]. In many cases [38] an exponent<br />
H > 1/2 has been found, indicating the existence of longrange<br />
correlation (persistence) in the data. It is <strong>to</strong> be noted,<br />
however, that the values of H computed using the traditional<br />
R/S-analysis, such as those quoted in [38], should be viewed<br />
with some caution, for this method has been shown [35] <strong>to</strong><br />
overestimate the value of H. In this sense, the DFA appears<br />
<strong>to</strong> give a more reliable estimates for H.<br />
<strong>An</strong> example of the DFA applied <strong>to</strong> the returns of the<br />
Ibovespa s<strong>to</strong>ck index is shown in Fig. 10 (upper curve). In<br />
this figure the upper straight line corresponds <strong>to</strong> the theoretical<br />
curve F H (τ) given in (149) with H = 0.6, and one<br />
sees an excellent agreement with the empirical data up <strong>to</strong><br />
τ ≃ 130 days. The fact that H > 0.5 thus indicates persistence<br />
in the Ibovespa returns. For τ > 130 the data deviate<br />
from the initial scaling behavior and cross over <strong>to</strong> a regime<br />
with a slope closer <strong>to</strong> 1/2, meaning that the Ibovespa looses<br />
its ‘memory’ after a period of about 6 months. Also shown<br />
in Fig. 10 is the corresponding F (τ) calculated for the shuffled<br />
Ibovespa returns. In this case we obtain an almost perfect<br />
scaling with H = 1/2, as expected, since the shuffling<br />
procedure tends <strong>to</strong> destroys any previously existing correlation.<br />
F(τ)<br />
10 1<br />
10 0<br />
ibovespa<br />
shuffled data<br />
H=0.6<br />
H=0.5<br />
10 1 10 2 10 3 10 4<br />
τ<br />
Figure 10. Fluctuation function F (τ) as a function of τ for the returns<br />
of the Ibovespa index (upper curve) and for the shuffled data<br />
(lower curve). The upper (lower) straight line gives the theoretical<br />
curve F H(τ) for H = 0.6 (H = 1/2).<br />
As already mentioned, the Hurst exponent has been<br />
calculated for many financial time series. In the case of<br />
s<strong>to</strong>ck indexes, the following interesting picture appears <strong>to</strong> be<br />
emerging from recent studies [40, 41, 35]: large and more<br />
developed markets, such as the New York and the London<br />
S<strong>to</strong>ck Exchanges, usually have H equal <strong>to</strong> (or slightly less<br />
than) 1/2, whereas less developed markets show a tendency<br />
<strong>to</strong> have H > 1/2. In other words, large markets seem indeed<br />
<strong>to</strong> be ‘efficient’ in the sense that H ≃ 1/2, whereas<br />
less developed markets tend <strong>to</strong> exhibit long-range correlation.<br />
A possible interpretation for this finding is that smaller<br />
markets are conceivably more prone <strong>to</strong> ‘correlated fluctuations’<br />
and perhaps more susceptible <strong>to</strong> being pushed around<br />
by aggressive inves<strong>to</strong>rs, which may explain in part a Hurst<br />
exponent greater than 1/2.